- Design
- Prospective, single-centre, single-arm feasibility study
- Population
- 98 English-speaking adults with a single-complaint urgent primary care appointment
- Primary outcome
- Safety stops, conversation quality ratings and patient and clinician experience
- Effect
- 0 safety stops; 1 hallucination; rated favourably on 87–100% of evaluator criteria; helpful to 33 of 44 doctors
In this single-centre, single-arm feasibility study, 98 adults booked for a single-complaint urgent primary care visit first had a text conversation with a diagnostic AI system up to five days beforehand. Physicians watched every conversation live, ready to stop it against prespecified safety criteria. The transcript and summary went to the treating doctor.
No conversation needed stopping. Supervisors recorded one hallucination and added clinical information in five conversations. Clinical evaluators rated the AI's conversations favourably on 87–100% of quality criteria. Of the 44 doctors who read the transcript before the visit and returned a survey, 33 found it helpful and 25 said it might have changed what they did.
This is early evidence of feasibility and acceptability, not of diagnostic accuracy or patient outcomes. Every conversation had a doctor watching, the sample is small, only English speakers took part, and the system's developer funded the study. It is a reason to follow the field, not to deploy an unsupervised tool.
- This study tested feasibility under live physician supervision; it did not test diagnostic accuracy or outcomes.
- One hallucination occurred in 98 conversations; any AI-generated history needs checking by the clinician.
- Treat an AI pre-visit summary as a draft history, not as findings.
- Ask patients whether they have used a chatbot before the visit; it shapes what they expect and report.
Why it matters
It is one of the first looks at an AI history-taker with real patients rather than actors.
Don't overread it
Every conversation was watched by a physician, so the safety record does not apply to unsupervised use.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for top clinical updates, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free